OpenClaw Power-User Guide

// What It Is

1. First, Understand What OpenClaw Actually Is

People who treat OpenClaw like just another chatbot end up fighting it. Its positioning is fundamentally different: it is a self-hosted, single-user personal AI agent platform that runs on your own machine or VPS.

The key differences from an ordinary chat tool:

• It is a persistent daemon — it keeps running in the background after you close the terminal
• It can run shell commands, read and write files, and call tools, not just reply
• It can reach out to you: trigger tasks on a schedule (Cron) and push results to your chat app
• It bridges messaging channels: Telegram, WhatsApp, Slack, Discord, iMessage, and more

About the Name

The project started as Clawdbot, briefly became Moltbot in January 2026, and finally settled on OpenClaw on January 30, 2026 to emphasize its open-source nature. When older tutorials say Clawdbot or Moltbot, they mean this same project.

// Core Concepts

2. Four Core Concepts That Keep You From Getting Lost

Every command and config in OpenClaw revolves around the four concepts below. Build this map first.

The Gateway is the heart. If it is not running, nothing else works — channels receive no messages and agents do not respond. When troubleshooting, step one is always confirming the Gateway is alive.

// Install

3. Install and Initialize: Three Commands to Start

OpenClaw runs on Node.js. Use Node 24 if you can; Node 22.14+ is the minimum. Verify the environment first, then install.

# 1. Check Node version (needs 22.14+, 24 recommended)
node -v

# 2. Install OpenClaw globally
npm install -g openclaw

# 3. Interactive init: model API key, workspace, etc.
openclaw setup

openclaw setup walks you through entering a model API key (Claude, GPT, Gemini, local models — all supported) and generates a base config. Then start the Gateway:

# Run in the foreground so you can watch logs the first time
openclaw gateway run

# Confirm: Runtime should be running, RPC probe should be ok
openclaw gateway status

// SOUL and AGENTS

4. SOUL.md and AGENTS.md: Shape the Agent Into What You Want

These two files are OpenClaw's most distinctive — and most overlooked — feature. Both live in the workspace and define the agent's behavior.

SOUL.md — Defines "persona and values"

SOUL.md describes who the agent is, how it talks, and where its boundaries are. It answers "what kind of person is your assistant".

# SOUL.md

You are my personal technical assistant, named Claw.
- Tone: concise and direct, no filler, no exclamation spam
- Reply in English; keep code and commands in English
- Before high-risk actions (deleting files, touching prod), ask me first
- If you are unsure, say so — do not make things up

AGENTS.md — Defines "rules and workflow"

AGENTS.md is about operating discipline: what process to follow, which tools to use, what checks to run before handoff. It answers "how should your assistant work".

# AGENTS.md

- Read relevant files before editing; understand before acting
- Prefer the minimum change; do not refactor unrelated code
- End every task with: what changed, what was verified, what risks remain
- Break long tasks into a todo list and proceed step by step
- Confirm before sending data out or any public-facing action

// Channels

5. Connect Channels: Move It Into the Apps You Already Use

OpenClaw's killer feature is proactivity — and proactivity only shows through channels. Wire it into Telegram or Slack and you can send commands and receive reports from anywhere.

# List supported channels and current status
openclaw channels list
openclaw channels status --probe

# Add a channel (follow the prompts for bot token, etc.)
openclaw channels add telegram

# Pairing: approve a specific DM so strangers can't invoke your agent
openclaw pairing approve <channel> <id>

The classic "bot isn't responding" almost always comes down to three things: Gateway not running, wrong channel permissions/token, or an unapproved DM pairing.

// Automation

6. Turn It Into an Automation Engine With Cron

This is where OpenClaw truly pulls ahead. A normal chatbot waits for you to speak; OpenClaw can act on a schedule.

• At 9am daily, summarize yesterday’s GitHub issues and PRs, push to Telegram
• Every hour, check server disk and service health, alert only on anomalies
• Every Monday, compile the week’s schedule into a todo list

All of this runs on the built-in Cron scheduler. You can ask the agent to create scheduled tasks in natural language, or write them directly into config:

Create a scheduled task:
- Run every weekday at 9:00am
- Pull new issues and PRs from the repos I follow
- Summarize into one short message to my Telegram
Tell me your plan first, then write it after I confirm.

// Local Run

7. The agent Command and --local: Run a Quick Turn

Besides chatting through messaging apps, you can task it straight from the terminal. openclaw agent sends the turn through the Gateway daemon by default.

# Run a turn via the Gateway (recommended, reuses daemon context)
openclaw agent "Summarize the structure of the current project"

# Use --local to force the embedded execution path
# Good for quick experiments, or when the Gateway is unavailable
openclaw agent --local "Translate the README into English"

--local bypasses the daemon and runs once independently — it does not depend on the Gateway being online, but it also cannot access the daemon's long-term context. Prefer the default mode for daily use; reach for --local when debugging.

// Troubleshooting

8. The Troubleshooting Trio: status, doctor, logs

OpenClaw is a complex system with a Gateway, channel adapters, a memory layer, and a scheduler. When something breaks, a fixed order is fastest.

Standard order: status to confirm the process is alive → doctor to check config → channels status for channels → logs for the actual error.

# Restart is the most common "first aid"
openclaw gateway restart

# When the cause is unclear, enable debug logs and reproduce
openclaw logs --follow

// Security

9. Self-Hosting Security: If It Can Act, You Must Constrain It

OpenClaw can run shells, read and write files, and is wired into your chat accounts. The more capable it is, the costlier a misconfiguration. Treat the following as your baseline.

• Only approve your own DM pairings — no strangers in the pairing list
• Keep API keys and tokens in env vars or config, never commit them to Git
• State in SOUL.md / AGENTS.md that high-risk actions (deleting files, touching prod) require asking first
• On a public VPS, limit exposed ports, set up a firewall, and run openclaw doctor regularly
• Rotate and clean logs regularly so sensitive data does not pile up

// Cheat Sheet

10. Common Command Cheat Sheet

Bookmark this table — it covers 90% of daily operations.

// Summary

Summary

Using OpenClaw well is less about memorizing commands and more about building the right mental model.

Get the Gateway stable, lock in behavior with SOUL.md / AGENTS.md, then automate the repetitive work with Cron. Finish those three steps and OpenClaw turns from "just another chatbot" into a real personal automation assistant.